On Thu 21-03-19 18:28:33, Oleg Nesterov wrote: > On 03/20, Michal Hocko wrote: > > > > So we need > > freezable_schedule_unsafe unsafe here to workaround the original problem > > and do not trigger the warning. > > Yes, but note the comment above freezable_schedule_unsafe() ;) > > that is why I didn't even try to suggest to use _unsafe() instead of reverting. Maybe I have misunderstood the discussion back then but we _knew_ that the lock is held and considered that an acceptable compromise to workaround the issue. And that is where the freezable_schedule_unsafe can be used AFAIU. > And to remind, until we solve the problem with cred_guard_mutex the freezer > can fail anyway although much less likely. Sure but this seems much less likely as the original patch seemed to resolve the failing suspend. So unless there is something to really handle cred_guard_mutex now, can we get back to the workaround and silence the lockdep splat and move on please? -- Michal Hocko SUSE Labs
